Hi, I'm Marc.

I build developer tools with AI models every day, and I pay for most of them myself: subscriptions, API credits, rented GPUs, and local models on my own machine.

Compute Crunch is where I write about what that is actually like. Which models hold up on real work and which only look good in benchmarks. Which harnesses waste your tokens. What happens when you give an agent too much access. And how the companies selling all of this treat the people paying for it.

The name is a description: we are in a compute crunch. Everyone wants more of it than exists, and the squeeze shows up in usage limits, "high demand" errors and support queues run by bots.

Everything here is first-hand and documented. When I quote a company, I have the email.

You can find me on X as @rustytensor.